A spatial–temporal graph structure automatic feedback learning system with tensor fusion and its application on engine RUL prediction

The prediction of remaining useful life (RUL) is a crucial technology in condition-based maintenance, playing a vital role in ensuring the reliability and safety of complex power equipment. To achieve accurate RUL prediction, we propose a graph structured automatic feedback learning model that optimizes graph topology for specific prediction tasks based on sensor time-series data. First, a graph structure learning method is developed to discover optimal graph connections according to the positional features of the nodes. Then, a parallel-route graph attention module is employed to extract spatial features from both the learned optimal graph and a designed graph constructed from prior knowledge. These two sets of features are then integrated into a high-order tensor, with richer multi-dimensional information preserved via a residual tensor fusion module. Additionally, a multi-head attention module based on the Transformer architecture is applied to process the tensor input and map the extracted features to the RUL. Finally, extensive experiments have been conducted on both the C-MAPSS dataset and a real-world engine test dataset to evaluate the effectiveness of the proposed method. The model consistently outperforms classical and advanced baselines across all sub-datasets. Ablation studies, complexity analysis, parameter sensitivity evaluations, and interpretability comparisons further validate its robustness and practical applicability.